Forward-Deployed AI Engineering
We build AIthat thinks.
A dedicated team of senior AI engineers.
Your CTO partner codes alongside you — from architecture to production.
The problem
Most companies are sitting on millions in untapped AI value.
They've tried chatbots that frustrate customers. Off-the-shelf tools that don't integrate. “AI-powered” features that are barely more than keyword search. The result: wasted budget, skeptical leadership, and a growing conviction that AI doesn't work.
AI works. The problem isn't the technology — it's the implementation. You need a true engineering partner: a dedicated team of senior AI specialists embedded in your stack, building for your data and your users. That's what we are.
Value delivered through AI at DBS Bank
AI systems shipped in 2026
Years building production AI systems
Our track record
What we build
Three disciplines. Infinite applications.
Every engagement is custom. A dedicated team of senior engineers partners directly with your team — from architecture decisions to production deployment.
01 — AI Agents & Employees
Your team, augmented.
We build AI agents that plug into your operations and run workflows end-to-end. They triage, decide, act, and learn from outcomes within guardrails you define.
Your team stops doing repetitive work and focuses on the calls that actually need a human.
Capabilities
- Customer operations agents that handle tickets end-to-end: triage, investigate, resolve, follow up
- Data analysis agents that monitor streams, surface anomalies, and generate reports automatically
- Workflow orchestration agents that coordinate across your CRM, databases, and communication tools
- Decision-support agents that synthesize complex information into actionable recommendations
02 — Mind Layer
AI with memory, personality, and depth.
Most AI is stateless—every conversation starts from zero. Mind Layer gives AI persistent memory, constructed affect for mood and emotional understanding, and adaptive personality. Model-agnostic by design: character state lives independently of any LLM, so you swap in better models the day they release without losing anything. Run production AI employees at 1/20th the model cost without sacrificing quality—because the engine handles the cognitive heavy lifting, not the LLM.
Users come back because the character remembers them. Retention grows over time instead of flattening. Upgrade the underlying model whenever you want without losing anything. No fine-tuning, no retraining.
Capabilities
- Persistent memory system—episodic, semantic, and procedural—that lets characters remember and evolve across sessions
- Affect system built on constructed emotion theory with allostatic mood regulation and co-regulatory relationship tracking
- Predictive personality framework that updates from social prediction errors while maintaining character consistency
- Model-agnostic architecture—memory, personality, and relational state persist across any LLM. No vendor lock-in, no retraining when better models appear
- 20x cost reduction—recursive LLM processing, efficient indexing, and multi-layer context assembly mean lightweight models achieve comparable output quality to frontier models. $0.10/conversation instead of $2.00, because the engine handles the cognitive heavy lifting
- Sub-200ms retrieval across thousands of memories for conversations that feel genuinely natural
03 — Custom AI Systems
Built for your exact requirements.
Off-the-shelf AI tools solve generic problems generically. We build bespoke AI employees, custom platforms, and tailored front-ends — integrated directly with your data and architecture. Whatever you need.
A fully custom AI system that fits your business like a glove — not a generic tool with workarounds.
Capabilities
- Bespoke AI employees built for your specific domain, workflows, and customer interactions
- Custom platforms and front-ends — your brand, your UX, your data, fully white-labeled
- Direct integration with your existing architecture — databases, CRMs, APIs, and internal systems
- End-to-end system architecture — from data pipeline design to model deployment and monitoring
- Enterprise-grade security with full data sovereignty, audit trails, and compliance controls
Why Sonzai
A partner, not a vendor.
A full team on your problem
You get a dedicated squad of ~10 senior AI engineers — not a solo contractor. Specialists in ML infrastructure, product, and integration, all focused on your project.
Your CTO codes your solution
Our CTO doesn’t just advise — he writes code on every project. The same architect who shipped $70M in production AI at DBS Bank is hands-on in your stack, personally.
Partners, not project managers
We don’t hand off deliverables and disappear. We stay embedded, iterate with your team, and treat your outcome as our own. You have our direct line, always.
The team
Two founders. Both in your stack.
Product and strategy on one side. AI architecture and engineering on the other. Both founders work hands-on in every project — you get their time, not a handoff to a junior team.
Ryan Foo
Ryan has shipped products across AI, gaming, and consumer tech. At Limit Break (raised $200M), he designed progression systems and economy mechanics for their flagship title. At Delphi Digital, he invested in consumer products and analyzed emerging technology markets.
- Ex-Game Designer, Limit Break (raised $200M)
- Ex-Consumer Investor, Delphi Digital
Nasrudin Salim
Nasrudin has architected and deployed production AI systems at Singapore’s largest financial institutions. As Head of MLOps at DBS Bank, he built the infrastructure that delivered $70M in annual value. At OCBC, he led Singapore’s first generative AI rollout as Lead AI Architect.
- Ex-VP Engineering, Almanak
- Ex-Head of MLOps, DBS Bank ($70M annual value)
- Ex-Lead AI Architect, OCBC Bank
- Led Singapore’s first Gen AI rollout
- Featured in The Straits Times
What we've built
From architecture to production.
We ship code, not recommendations. Here's what we've built and what we're building.
Sonzai Mind Layer
Model-agnostic AI employee system with persistent memory, affect modeling, and personality adaptation. Character state lives independently of any LLM, so you swap providers without losing anything. 20x cost reduction through orchestration: lightweight models match frontier output quality because the engine handles the cognitive work. Sub-200ms response times in production.
Pocket Souls
AI companion app where characters remember you, evolve with you, and develop real personality over time. Built on Mind Layer. Our own product, proving the tech works at consumer scale.
AI Employee Systems
Autonomous AI agent workflows for enterprise operations. Customer support, data analysis, lead processing. End-to-end automation within defined guardrails.
How we work
Embedded partners from day one.
Every engagement follows the same rhythm: a dedicated team embedded in your stack, the CTO coding alongside you, and founders at every decision point.
Discovery
We learn your business context, existing systems, and goals. Getting architecture right early saves months later.
1–2 weeksArchitecture
System architecture, data flows, integration points. Every decision shaped by your constraints and scale needs.
1–2 weeksBuild
Iterative development with regular checkpoints. First version is intentionally minimal, focused on validating assumptions with real users.
4–12 weeksDeploy & Evolve
Production launch with monitoring and feedback loops. The system improves over time. We stay involved as long as you need us.
OngoingExample solutions we've built
Every business is different. Here are examples of what bespoke AI employees look like in practice.
Lead Qualification
AI employee that processes, scores, and nurtures leads 24/7
Travel Agencies
Proactive itinerary intelligence with persistent customer memory
AI Companions
Companion apps with adaptive personality and persistent memory
AI Tutors
Tutors that remember every student and adapt teaching style
These are examples — we build whatever your business needs. Talk to us about yours.
Latest thinking
From the blog.
Free consultation.
Free proposal.
Tell us what you need. We'll pair you with our CTO and a dedicated engineering team, scope a solution specific to your business, and deliver a custom proposal — at no cost. No commitment until you're ready.